川东石炭系新区大多数气藏尚未进入或结束评价阶段，资料不足而不具备进行数值模拟编制早期开发方案的条件。为了高速高效开发气藏及编制矿区上产规划，就必须选择适当的方法进行新区早期动态预测。本文正是基于此，特提出递推滚动ＢＰ神经网络预测模型。利用老区不同类型气藏的已建模型，对新区不同采气速度下，稳产期和递减期地层压力的变化规律、最终采出程度的大小进行了预测。预测精度与传统的数值模拟方法相当，更具有简便快速的优点。 Most of the gas reservoirs in the East Sichuan Carboniferous New Area have not yet entered or concluded the evaluation stage, and the data are not sufficient and do not have the conditions for carrying out numerical simulation to formulate the early development plan. In order to develop gas reservoirs at high speed and high efficiency, and to plan production up-production in the mining area, an appropriate method must be selected for early dynamic prediction of the new area. Based on this, this paper proposes a recursive rolling BP neural network prediction model. Based on the established models of different types of gas reservoirs in the old area, the variation laws of formation pressure and the degree of final recovery are predicted under different gas production rates in the new area. The prediction accuracy is equivalent to the traditional numerical simulation method, moreover, it has the advantages of simplicity and speed.